From Product Hunt
Agent infrastructure day on PH. MCP servers, agent debuggers, and computer use tools dominated the top launches.
Theneo — API Management for Humans and Agents
All-in-one platform to design, document, manage, and observe APIs. Generates production-grade MCP servers from your repos, audits APIs for agent-readiness, and auto-generates docs that AI agents can actually parse.
Notion Developer Platform — Build on Notion, Not Just Inside It
Notion goes from workspace to agent infrastructure. New CLI, hosted Workers for custom logic, upgraded MCP, and External Agents API that lets Claude Code, Codex, and Cursor operate as first-class collaborators inside your workspace.
Raindrop Workshop — Local Debugger for AI Agents
Free, open-source, local-first agent debugger. Traces stream token-by-token in real time. Claude Code reads traces over MCP, writes evals, fixes bugs, and re-runs — a self-healing agent loop. Supports every major agent SDK.
Resend Automations — Event-Driven Email Flows
Build email workflows with triggers, conditions, delays, and full run visibility. Already at 518K+ automation runs since launch week. Top-tier DX from the Resend team.
Asteroid — Computer Use AI Agents in Seconds
Build browser, Linux, and Windows computer-use agents in minutes. Meta-agent Astro builds agents and writes scripts as it goes. 150K+ executions last month across EHRs, insurance portals, and Citrix. HIPAA/SOC2 certified. YC W25.
Edit Mind — Search Videos Like You'd Describe to a Friend
100% local video search for content creators. Transcribes footage on-device, analyzes frames (faces, objects, scenes, text), and indexes everything for natural language queries. No cloud, no API bills.
From Reddit
Mythos Preview helped build a macOS kernel exploit in 5 days. Figure's robots crossed 30 hours of autonomous work. Codex goes mobile.
Security firm Calif used Anthropic's Mythos Preview to build the first public macOS kernel memory corruption exploit targeting M5 silicon. A data-only local privilege escalation chain going from unprivileged user to root shell, bypassing Apple's Memory Integrity Enforcement.
Bruce Dang found the bugs April 25th, Dion Blazakis joined April 27th, and by May 1st they had a working exploit. A 55-page report will be published after Apple patches. Confirms Mythos's cybersecurity capabilities are very real.
Figure 03 Robots Cross 30 Hours of Autonomous Work Without Failure
Three Figure 03 robots — Gary, Bob, and Frank — completed 30+ hours of nonstop package sorting on a surprise YouTube livestream. Fully autonomous via onboard Helix-02 neural network, no teleoperation. Sorted 28,000+ packages at near-human speed. CEO Brett Adcock called it "uncharted territory."
OpenAI Ships Codex Mobile — Start Work from Your Phone
Codex now runs in the ChatGPT mobile app on iOS and Android. Start new work, review outputs, steer execution, and approve next steps from your phone while Codex keeps running on your laptop or devbox. Also launched a Windows sandbox for secure Codex execution.
Poetiq: Recursive Self-Improvement Delivers New SOTA Coding Performance
Poetiq's Meta-System constructs and optimizes its own coding harnesses from scratch. Improved Gemini 3.1 Pro from 78.6% to 90.9% on LiveCodeBench Pro, overtaking GPT-5.5. Applied the same harness to GPT-5.5 itself and pushed it to 93.9%. Model-agnostic — no fine-tuning needed.
RTX 5090 Price Hike Coming — GDDR7 Costs Rising
Nvidia reportedly preparing a price increase on the RTX 5090 due to rising GDDR7 memory costs. It's the only GPU tier not dropping in price across EU retailers.
Nvidia Releases NVFP4 Quantized Kimi 2.6 and 2.5
Nvidia's official FP4 quantizations of Moonshot AI's Kimi K2.6, ready for commercial use. Minimal accuracy loss from FP4 compression on key benchmarks.
OpenAI Hit with Class-Action Privacy Lawsuit Over ChatGPT Data Sharing
A California resident filed suit claiming OpenAI sent ChatGPT queries to Google and Meta via tracking pixels. Could reshape AI data-privacy rules.
From Reddit
Anima ships its v1.0 final release, and the LTX 2.3 ecosystem gets a proper timeline editor.
Anima Base v1.0 Released — ComfyUI Grant Image Model Goes Final
Circlestone Labs ships the v1.0 final release of Anima, the 2B image model backed by ComfyUI's $1M Open AI Grant. Major improvements over Preview 3 in natural language comprehension and character consistency. LoRAs trained on earlier previews still work. Available on Civitai and HuggingFace.
LTX Director — All-In-One Timeline Editor for LTX 2.3 Videos
Free ComfyUI node that brings timeline editing to LTX 2.3 video generation: I2V, T2V, first/last frame, prompt relay, custom audio, and multi-shot composition. Built in 6 days of 16-hour sessions vibe coding with Gemini. Unlocks the full potential of LTX 2.3.
AsymFLUX.2-klein-9B — Pixel-Space Text-to-Image Model
Novel asymmetric flow model fine-tuned from FLUX.2-klein-base-9B. Generates images directly in pixel space instead of latent space. Paper and weights available.
Qwen-Image-VAE-2.0 — High-Compression VAE Suite
Alibaba releases new VAE family with significant advances in both reconstruction fidelity and diffusability. Global skip connections and improved architecture for high compression ratios.
From Reddit
A Screen Studio competitor with 3D renders, an open-source live wallpaper app, and a clipboard manager that finally does workspaces.
ShotGlass — Screen Recorder with Cinematic 3D Demos for Mac
Record your screen and instantly render it with 3D MacBook frames, automatic zoom, and background replacement. One-time purchase, no subscription. Competes directly with Screen Studio at a fraction of the cost.
Luviosa — Open-Source macOS Live Wallpaper App
Native SwiftUI app for animated wallpapers on Mac. Browse curated wallpapers by category, import your own videos/GIFs, apply to multiple displays. Menu bar only — no Dock clutter. Fully open source on GitHub.
OpenClaw OS — Turn One-Off Chats into Persistent Apps
OpenClaw goes beyond chat with a proper agent OS: named sessions, persistent apps, scheduled cron jobs, generative UI, and multi-agent workspaces. Your agent builds a dashboard, and it stays running. No more losing work in scrollback.